The generated metadata for rayleigh_benard_uniform gives dataset_name: rayleigh_benard, and both supernova_explosion_64 and _128 give supernova_explosion. A hosted supernova_explosion_64 header has the family name too. So the registry key and the stored name are different, and I could not tell which one is meant to be the canonical one.
Nothing in the loader catches this. WellDataset only checks that the files agree with each other. name_override sets dataset.dataset_name after metadata is built, so dataset.metadata.dataset_name keeps the stored value. Benchmark metric keys and plot paths use the metadata name, so I would expect supernova_explosion_64 and _128 to log under the same key.
Is the registry key meant to be the canonical public name, in WellMetadata and in logging? Or are the shared stored names intentional? If you tell me which one it should be, I can write the change and the regression tests.
The generated metadata for
rayleigh_benard_uniformgivesdataset_name: rayleigh_benard, and bothsupernova_explosion_64and_128givesupernova_explosion. A hostedsupernova_explosion_64header has the family name too. So the registry key and the stored name are different, and I could not tell which one is meant to be the canonical one.Nothing in the loader catches this.
WellDatasetonly checks that the files agree with each other.name_overridesetsdataset.dataset_nameaftermetadatais built, sodataset.metadata.dataset_namekeeps the stored value. Benchmark metric keys and plot paths use the metadata name, so I would expectsupernova_explosion_64and_128to log under the same key.Is the registry key meant to be the canonical public name, in
WellMetadataand in logging? Or are the shared stored names intentional? If you tell me which one it should be, I can write the change and the regression tests.